Research On Method Of Image Segmentation Based On Active Contour Model

The technology of image segmentation means extracting interested objects from images and it is one of key issues in image processing. This dissertation presented some studies concentrated in the following four topics:image segmentation based on geodesic active contour model, image segmentation based on C-V model, image segmentation based on mean shift and image segmentation based on normalized cuts.For geodesic active contour, studies were focused on its force field design. Gradient vector flow is one of the most widely used force fields in the field of active contour models. In spite of its superior performance, it is still difficult for gradient vector flow geodesic active contour to enter into special shape concavities and get rid of noise disturbance. In order to solve the above problems, author put forward inertia force field for geodesic active contours and then further put forward gradient vector flow and inertia geodesic active contour through incorporating gradient vector flow and inertia into geodesic active contours. Before the author put forward inertia force, all active contours evolved in non-inertial system. Inertia force field transferred the evolution of geodesic active contours into inertial system and therefore they have the ability to begin to evolve with initial speed. With the help of inertia and initial speed, geodesic active contours can enter into special shape concavities easily. More importantly, inertia and initial speed make contour curves more tolerant toward initial curves.As a region-based active contour model, C-V model can detect fuzzy and discrete boundaries and shows better adaptability to noise disturbance. However, for the images which can not be fitted by separatrix smooth two-value functions, segmentation effects of C-V model may not be ideal. The author integrated region force of C-V model with gradient vector flow and use a factor to adjust the weight of region force in this coupled force field. This factor can change with the change of contour curves’ internal area. Using this coupled force field as external force, the author put forward gradient vector flow geodesic active contour aided region force. In this model, the effect of region force decreases with the increase of the effect of gradient vector flow. It can not only overcome noise disturbance, but also adapt complex background. The author used experiments to verify its ability from several aspects.Image segmentation technique based on mean shift is building on non-parameter density estimation. The author integrated this method with modified region force of C-V model and put forward adapting complex background C-V model. This model has better adaptability to complex background and overcomes the defect that image segmentation technique based on mean shift always segment out irregular zigzag boundaries.Normalized cut is an image segmentation technique building on global opti-mization and can give hierarchical description of images’ global impression. Through integrating pixon image model with normalized cut, the author put forward normalized cut based on pixon image model. It changes the objects that normalized cut segments from massive pixels in the images to less pixons in pixon image modesl. Under the condition that segmen- tation effects keep almost invariant, pixon-based normalized cut accelerated the segmentation speed and reduced the consuming of memory.